How Trakkr Spots a Breakdown Before It Happens

Predicting an elevator failure sounds like it should need expensive sensors and live monitoring hardware bolted to every shaft. Trakkr does it with none of that — just the public availability records the MTA has quietly been publishing for years.

The simple idea

Elevators don't break down out of nowhere. They break down in patterns. A unit that's going to fail next month almost always had more outages last month than usual. The warning window between the first hiccup and a full breakdown is real, and it's measurable. Trakkr is built to measure it.

The bet was simple: years of past availability data hold enough warning signs to predict future breakdowns, without needing fancy sensors or live equipment feeds. On 2025 records, Trakkr correctly flagged about 96 out of every 100 elevators that actually went down. The bet held up.

What Trakkr actually looks at

For every elevator and escalator in the system, Trakkr looks at recent history at several time scales:

• How often it's been working — over the last month, three months, and six months.
• How often it's been out — both planned and unplanned outages.
• Whether it has a history of trapping riders inside. Units that have stranded passengers before are the ones we watch closest.
• How long it's been since a major repair. Recently serviced units behave differently than ones overdue for attention.
• What kind of unit it is — elevator or escalator — and where it sits in the network.

None of these signals are exotic. They're things any rider could spot if they were tracking 689 units across the city by hand. Trakkr just does the tracking.

Two forecasts, not one

Elevators and escalators fail in different ways. Elevator outages tend to be longer and carry entrapment risk. Escalators fail more often but recover faster. Mixing the two into a single forecast hides what makes each kind dangerous, so Trakkr keeps them separate.

Why Trakkr leans toward false alarms over missed alarms

No forecast is perfect. Trakkr is tuned to err on the side of warning you about a unit that ends up working fine, rather than missing a unit that actually breaks down. The math on that trade-off is straightforward:

• If we warn you about an elevator that ends up working, you check an alternative route. Cost: a few seconds.
• If we miss an elevator that breaks down, you're stuck on a platform with a stroller, a wheelchair, or a tight connection. Cost: a real disruption to your day.

That asymmetry is built into how Trakkr decides when to flag a unit.

For brand-new units with no history yet, Trakkr fills in the gaps using averages from similar units in the same borough — so a fresh elevator doesn't get an automatic "safe" pass just because it has no track record.